4 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
5 * Swap reorganised 29.12.95, Stephen Tweedie
9 #include <linux/hugetlb.h>
10 #include <linux/mman.h>
11 #include <linux/slab.h>
12 #include <linux/kernel_stat.h>
13 #include <linux/swap.h>
14 #include <linux/vmalloc.h>
15 #include <linux/pagemap.h>
16 #include <linux/namei.h>
17 #include <linux/shmem_fs.h>
18 #include <linux/blkdev.h>
19 #include <linux/random.h>
20 #include <linux/writeback.h>
21 #include <linux/proc_fs.h>
22 #include <linux/seq_file.h>
23 #include <linux/init.h>
24 #include <linux/ksm.h>
25 #include <linux/rmap.h>
26 #include <linux/security.h>
27 #include <linux/backing-dev.h>
28 #include <linux/mutex.h>
29 #include <linux/capability.h>
30 #include <linux/syscalls.h>
31 #include <linux/memcontrol.h>
32 #include <linux/poll.h>
33 #include <linux/oom.h>
35 #include <asm/pgtable.h>
36 #include <asm/tlbflush.h>
37 #include <linux/swapops.h>
38 #include <linux/page_cgroup.h>
40 static bool swap_count_continued(struct swap_info_struct
*, pgoff_t
,
42 static void free_swap_count_continuations(struct swap_info_struct
*);
43 static sector_t
map_swap_entry(swp_entry_t
, struct block_device
**);
45 static DEFINE_SPINLOCK(swap_lock
);
46 static unsigned int nr_swapfiles
;
48 long total_swap_pages
;
49 static int least_priority
;
51 static const char Bad_file
[] = "Bad swap file entry ";
52 static const char Unused_file
[] = "Unused swap file entry ";
53 static const char Bad_offset
[] = "Bad swap offset entry ";
54 static const char Unused_offset
[] = "Unused swap offset entry ";
56 static struct swap_list_t swap_list
= {-1, -1};
58 static struct swap_info_struct
*swap_info
[MAX_SWAPFILES
];
60 static DEFINE_MUTEX(swapon_mutex
);
62 static DECLARE_WAIT_QUEUE_HEAD(proc_poll_wait
);
63 /* Activity counter to indicate that a swapon or swapoff has occurred */
64 static atomic_t proc_poll_event
= ATOMIC_INIT(0);
66 static inline unsigned char swap_count(unsigned char ent
)
68 return ent
& ~SWAP_HAS_CACHE
; /* may include SWAP_HAS_CONT flag */
71 /* returns 1 if swap entry is freed */
73 __try_to_reclaim_swap(struct swap_info_struct
*si
, unsigned long offset
)
75 swp_entry_t entry
= swp_entry(si
->type
, offset
);
79 page
= find_get_page(&swapper_space
, entry
.val
);
83 * This function is called from scan_swap_map() and it's called
84 * by vmscan.c at reclaiming pages. So, we hold a lock on a page, here.
85 * We have to use trylock for avoiding deadlock. This is a special
86 * case and you should use try_to_free_swap() with explicit lock_page()
87 * in usual operations.
89 if (trylock_page(page
)) {
90 ret
= try_to_free_swap(page
);
93 page_cache_release(page
);
98 * swapon tell device that all the old swap contents can be discarded,
99 * to allow the swap device to optimize its wear-levelling.
101 static int discard_swap(struct swap_info_struct
*si
)
103 struct swap_extent
*se
;
104 sector_t start_block
;
108 /* Do not discard the swap header page! */
109 se
= &si
->first_swap_extent
;
110 start_block
= (se
->start_block
+ 1) << (PAGE_SHIFT
- 9);
111 nr_blocks
= ((sector_t
)se
->nr_pages
- 1) << (PAGE_SHIFT
- 9);
113 err
= blkdev_issue_discard(si
->bdev
, start_block
,
114 nr_blocks
, GFP_KERNEL
, 0);
120 list_for_each_entry(se
, &si
->first_swap_extent
.list
, list
) {
121 start_block
= se
->start_block
<< (PAGE_SHIFT
- 9);
122 nr_blocks
= (sector_t
)se
->nr_pages
<< (PAGE_SHIFT
- 9);
124 err
= blkdev_issue_discard(si
->bdev
, start_block
,
125 nr_blocks
, GFP_KERNEL
, 0);
131 return err
; /* That will often be -EOPNOTSUPP */
135 * swap allocation tell device that a cluster of swap can now be discarded,
136 * to allow the swap device to optimize its wear-levelling.
138 static void discard_swap_cluster(struct swap_info_struct
*si
,
139 pgoff_t start_page
, pgoff_t nr_pages
)
141 struct swap_extent
*se
= si
->curr_swap_extent
;
142 int found_extent
= 0;
145 struct list_head
*lh
;
147 if (se
->start_page
<= start_page
&&
148 start_page
< se
->start_page
+ se
->nr_pages
) {
149 pgoff_t offset
= start_page
- se
->start_page
;
150 sector_t start_block
= se
->start_block
+ offset
;
151 sector_t nr_blocks
= se
->nr_pages
- offset
;
153 if (nr_blocks
> nr_pages
)
154 nr_blocks
= nr_pages
;
155 start_page
+= nr_blocks
;
156 nr_pages
-= nr_blocks
;
159 si
->curr_swap_extent
= se
;
161 start_block
<<= PAGE_SHIFT
- 9;
162 nr_blocks
<<= PAGE_SHIFT
- 9;
163 if (blkdev_issue_discard(si
->bdev
, start_block
,
164 nr_blocks
, GFP_NOIO
, 0))
169 se
= list_entry(lh
, struct swap_extent
, list
);
173 static int wait_for_discard(void *word
)
179 #define SWAPFILE_CLUSTER 256
180 #define LATENCY_LIMIT 256
182 static unsigned long scan_swap_map(struct swap_info_struct
*si
,
185 unsigned long offset
;
186 unsigned long scan_base
;
187 unsigned long last_in_cluster
= 0;
188 int latency_ration
= LATENCY_LIMIT
;
189 int found_free_cluster
= 0;
192 * We try to cluster swap pages by allocating them sequentially
193 * in swap. Once we've allocated SWAPFILE_CLUSTER pages this
194 * way, however, we resort to first-free allocation, starting
195 * a new cluster. This prevents us from scattering swap pages
196 * all over the entire swap partition, so that we reduce
197 * overall disk seek times between swap pages. -- sct
198 * But we do now try to find an empty cluster. -Andrea
199 * And we let swap pages go all over an SSD partition. Hugh
202 si
->flags
+= SWP_SCANNING
;
203 scan_base
= offset
= si
->cluster_next
;
205 if (unlikely(!si
->cluster_nr
--)) {
206 if (si
->pages
- si
->inuse_pages
< SWAPFILE_CLUSTER
) {
207 si
->cluster_nr
= SWAPFILE_CLUSTER
- 1;
210 if (si
->flags
& SWP_DISCARDABLE
) {
212 * Start range check on racing allocations, in case
213 * they overlap the cluster we eventually decide on
214 * (we scan without swap_lock to allow preemption).
215 * It's hardly conceivable that cluster_nr could be
216 * wrapped during our scan, but don't depend on it.
218 if (si
->lowest_alloc
)
220 si
->lowest_alloc
= si
->max
;
221 si
->highest_alloc
= 0;
223 spin_unlock(&swap_lock
);
226 * If seek is expensive, start searching for new cluster from
227 * start of partition, to minimize the span of allocated swap.
228 * But if seek is cheap, search from our current position, so
229 * that swap is allocated from all over the partition: if the
230 * Flash Translation Layer only remaps within limited zones,
231 * we don't want to wear out the first zone too quickly.
233 if (!(si
->flags
& SWP_SOLIDSTATE
))
234 scan_base
= offset
= si
->lowest_bit
;
235 last_in_cluster
= offset
+ SWAPFILE_CLUSTER
- 1;
237 /* Locate the first empty (unaligned) cluster */
238 for (; last_in_cluster
<= si
->highest_bit
; offset
++) {
239 if (si
->swap_map
[offset
])
240 last_in_cluster
= offset
+ SWAPFILE_CLUSTER
;
241 else if (offset
== last_in_cluster
) {
242 spin_lock(&swap_lock
);
243 offset
-= SWAPFILE_CLUSTER
- 1;
244 si
->cluster_next
= offset
;
245 si
->cluster_nr
= SWAPFILE_CLUSTER
- 1;
246 found_free_cluster
= 1;
249 if (unlikely(--latency_ration
< 0)) {
251 latency_ration
= LATENCY_LIMIT
;
255 offset
= si
->lowest_bit
;
256 last_in_cluster
= offset
+ SWAPFILE_CLUSTER
- 1;
258 /* Locate the first empty (unaligned) cluster */
259 for (; last_in_cluster
< scan_base
; offset
++) {
260 if (si
->swap_map
[offset
])
261 last_in_cluster
= offset
+ SWAPFILE_CLUSTER
;
262 else if (offset
== last_in_cluster
) {
263 spin_lock(&swap_lock
);
264 offset
-= SWAPFILE_CLUSTER
- 1;
265 si
->cluster_next
= offset
;
266 si
->cluster_nr
= SWAPFILE_CLUSTER
- 1;
267 found_free_cluster
= 1;
270 if (unlikely(--latency_ration
< 0)) {
272 latency_ration
= LATENCY_LIMIT
;
277 spin_lock(&swap_lock
);
278 si
->cluster_nr
= SWAPFILE_CLUSTER
- 1;
279 si
->lowest_alloc
= 0;
283 if (!(si
->flags
& SWP_WRITEOK
))
285 if (!si
->highest_bit
)
287 if (offset
> si
->highest_bit
)
288 scan_base
= offset
= si
->lowest_bit
;
290 /* reuse swap entry of cache-only swap if not busy. */
291 if (vm_swap_full() && si
->swap_map
[offset
] == SWAP_HAS_CACHE
) {
293 spin_unlock(&swap_lock
);
294 swap_was_freed
= __try_to_reclaim_swap(si
, offset
);
295 spin_lock(&swap_lock
);
296 /* entry was freed successfully, try to use this again */
299 goto scan
; /* check next one */
302 if (si
->swap_map
[offset
])
305 if (offset
== si
->lowest_bit
)
307 if (offset
== si
->highest_bit
)
310 if (si
->inuse_pages
== si
->pages
) {
311 si
->lowest_bit
= si
->max
;
314 si
->swap_map
[offset
] = usage
;
315 si
->cluster_next
= offset
+ 1;
316 si
->flags
-= SWP_SCANNING
;
318 if (si
->lowest_alloc
) {
320 * Only set when SWP_DISCARDABLE, and there's a scan
321 * for a free cluster in progress or just completed.
323 if (found_free_cluster
) {
325 * To optimize wear-levelling, discard the
326 * old data of the cluster, taking care not to
327 * discard any of its pages that have already
328 * been allocated by racing tasks (offset has
329 * already stepped over any at the beginning).
331 if (offset
< si
->highest_alloc
&&
332 si
->lowest_alloc
<= last_in_cluster
)
333 last_in_cluster
= si
->lowest_alloc
- 1;
334 si
->flags
|= SWP_DISCARDING
;
335 spin_unlock(&swap_lock
);
337 if (offset
< last_in_cluster
)
338 discard_swap_cluster(si
, offset
,
339 last_in_cluster
- offset
+ 1);
341 spin_lock(&swap_lock
);
342 si
->lowest_alloc
= 0;
343 si
->flags
&= ~SWP_DISCARDING
;
345 smp_mb(); /* wake_up_bit advises this */
346 wake_up_bit(&si
->flags
, ilog2(SWP_DISCARDING
));
348 } else if (si
->flags
& SWP_DISCARDING
) {
350 * Delay using pages allocated by racing tasks
351 * until the whole discard has been issued. We
352 * could defer that delay until swap_writepage,
353 * but it's easier to keep this self-contained.
355 spin_unlock(&swap_lock
);
356 wait_on_bit(&si
->flags
, ilog2(SWP_DISCARDING
),
357 wait_for_discard
, TASK_UNINTERRUPTIBLE
);
358 spin_lock(&swap_lock
);
361 * Note pages allocated by racing tasks while
362 * scan for a free cluster is in progress, so
363 * that its final discard can exclude them.
365 if (offset
< si
->lowest_alloc
)
366 si
->lowest_alloc
= offset
;
367 if (offset
> si
->highest_alloc
)
368 si
->highest_alloc
= offset
;
374 spin_unlock(&swap_lock
);
375 while (++offset
<= si
->highest_bit
) {
376 if (!si
->swap_map
[offset
]) {
377 spin_lock(&swap_lock
);
380 if (vm_swap_full() && si
->swap_map
[offset
] == SWAP_HAS_CACHE
) {
381 spin_lock(&swap_lock
);
384 if (unlikely(--latency_ration
< 0)) {
386 latency_ration
= LATENCY_LIMIT
;
389 offset
= si
->lowest_bit
;
390 while (++offset
< scan_base
) {
391 if (!si
->swap_map
[offset
]) {
392 spin_lock(&swap_lock
);
395 if (vm_swap_full() && si
->swap_map
[offset
] == SWAP_HAS_CACHE
) {
396 spin_lock(&swap_lock
);
399 if (unlikely(--latency_ration
< 0)) {
401 latency_ration
= LATENCY_LIMIT
;
404 spin_lock(&swap_lock
);
407 si
->flags
-= SWP_SCANNING
;
411 swp_entry_t
get_swap_page(void)
413 struct swap_info_struct
*si
;
418 spin_lock(&swap_lock
);
419 if (nr_swap_pages
<= 0)
423 for (type
= swap_list
.next
; type
>= 0 && wrapped
< 2; type
= next
) {
424 si
= swap_info
[type
];
427 (!wrapped
&& si
->prio
!= swap_info
[next
]->prio
)) {
428 next
= swap_list
.head
;
432 if (!si
->highest_bit
)
434 if (!(si
->flags
& SWP_WRITEOK
))
437 swap_list
.next
= next
;
438 /* This is called for allocating swap entry for cache */
439 offset
= scan_swap_map(si
, SWAP_HAS_CACHE
);
441 spin_unlock(&swap_lock
);
442 return swp_entry(type
, offset
);
444 next
= swap_list
.next
;
449 spin_unlock(&swap_lock
);
450 return (swp_entry_t
) {0};
453 /* The only caller of this function is now susupend routine */
454 swp_entry_t
get_swap_page_of_type(int type
)
456 struct swap_info_struct
*si
;
459 spin_lock(&swap_lock
);
460 si
= swap_info
[type
];
461 if (si
&& (si
->flags
& SWP_WRITEOK
)) {
463 /* This is called for allocating swap entry, not cache */
464 offset
= scan_swap_map(si
, 1);
466 spin_unlock(&swap_lock
);
467 return swp_entry(type
, offset
);
471 spin_unlock(&swap_lock
);
472 return (swp_entry_t
) {0};
475 static struct swap_info_struct
*swap_info_get(swp_entry_t entry
)
477 struct swap_info_struct
*p
;
478 unsigned long offset
, type
;
482 type
= swp_type(entry
);
483 if (type
>= nr_swapfiles
)
486 if (!(p
->flags
& SWP_USED
))
488 offset
= swp_offset(entry
);
489 if (offset
>= p
->max
)
491 if (!p
->swap_map
[offset
])
493 spin_lock(&swap_lock
);
497 printk(KERN_ERR
"swap_free: %s%08lx\n", Unused_offset
, entry
.val
);
500 printk(KERN_ERR
"swap_free: %s%08lx\n", Bad_offset
, entry
.val
);
503 printk(KERN_ERR
"swap_free: %s%08lx\n", Unused_file
, entry
.val
);
506 printk(KERN_ERR
"swap_free: %s%08lx\n", Bad_file
, entry
.val
);
511 static unsigned char swap_entry_free(struct swap_info_struct
*p
,
512 swp_entry_t entry
, unsigned char usage
)
514 unsigned long offset
= swp_offset(entry
);
516 unsigned char has_cache
;
518 count
= p
->swap_map
[offset
];
519 has_cache
= count
& SWAP_HAS_CACHE
;
520 count
&= ~SWAP_HAS_CACHE
;
522 if (usage
== SWAP_HAS_CACHE
) {
523 VM_BUG_ON(!has_cache
);
525 } else if (count
== SWAP_MAP_SHMEM
) {
527 * Or we could insist on shmem.c using a special
528 * swap_shmem_free() and free_shmem_swap_and_cache()...
531 } else if ((count
& ~COUNT_CONTINUED
) <= SWAP_MAP_MAX
) {
532 if (count
== COUNT_CONTINUED
) {
533 if (swap_count_continued(p
, offset
, count
))
534 count
= SWAP_MAP_MAX
| COUNT_CONTINUED
;
536 count
= SWAP_MAP_MAX
;
542 mem_cgroup_uncharge_swap(entry
);
544 usage
= count
| has_cache
;
545 p
->swap_map
[offset
] = usage
;
547 /* free if no reference */
549 struct gendisk
*disk
= p
->bdev
->bd_disk
;
550 if (offset
< p
->lowest_bit
)
551 p
->lowest_bit
= offset
;
552 if (offset
> p
->highest_bit
)
553 p
->highest_bit
= offset
;
554 if (swap_list
.next
>= 0 &&
555 p
->prio
> swap_info
[swap_list
.next
]->prio
)
556 swap_list
.next
= p
->type
;
559 if ((p
->flags
& SWP_BLKDEV
) &&
560 disk
->fops
->swap_slot_free_notify
)
561 disk
->fops
->swap_slot_free_notify(p
->bdev
, offset
);
568 * Caller has made sure that the swapdevice corresponding to entry
569 * is still around or has not been recycled.
571 void swap_free(swp_entry_t entry
)
573 struct swap_info_struct
*p
;
575 p
= swap_info_get(entry
);
577 swap_entry_free(p
, entry
, 1);
578 spin_unlock(&swap_lock
);
583 * Called after dropping swapcache to decrease refcnt to swap entries.
585 void swapcache_free(swp_entry_t entry
, struct page
*page
)
587 struct swap_info_struct
*p
;
590 p
= swap_info_get(entry
);
592 count
= swap_entry_free(p
, entry
, SWAP_HAS_CACHE
);
594 mem_cgroup_uncharge_swapcache(page
, entry
, count
!= 0);
595 spin_unlock(&swap_lock
);
600 * How many references to page are currently swapped out?
601 * This does not give an exact answer when swap count is continued,
602 * but does include the high COUNT_CONTINUED flag to allow for that.
604 int page_swapcount(struct page
*page
)
607 struct swap_info_struct
*p
;
610 entry
.val
= page_private(page
);
611 p
= swap_info_get(entry
);
613 count
= swap_count(p
->swap_map
[swp_offset(entry
)]);
614 spin_unlock(&swap_lock
);
620 * We can write to an anon page without COW if there are no other references
621 * to it. And as a side-effect, free up its swap: because the old content
622 * on disk will never be read, and seeking back there to write new content
623 * later would only waste time away from clustering.
625 int reuse_swap_page(struct page
*page
)
629 VM_BUG_ON(!PageLocked(page
));
630 if (unlikely(PageKsm(page
)))
632 count
= page_mapcount(page
);
633 if (count
<= 1 && PageSwapCache(page
)) {
634 count
+= page_swapcount(page
);
635 if (count
== 1 && !PageWriteback(page
)) {
636 delete_from_swap_cache(page
);
644 * If swap is getting full, or if there are no more mappings of this page,
645 * then try_to_free_swap is called to free its swap space.
647 int try_to_free_swap(struct page
*page
)
649 VM_BUG_ON(!PageLocked(page
));
651 if (!PageSwapCache(page
))
653 if (PageWriteback(page
))
655 if (page_swapcount(page
))
659 * Once hibernation has begun to create its image of memory,
660 * there's a danger that one of the calls to try_to_free_swap()
661 * - most probably a call from __try_to_reclaim_swap() while
662 * hibernation is allocating its own swap pages for the image,
663 * but conceivably even a call from memory reclaim - will free
664 * the swap from a page which has already been recorded in the
665 * image as a clean swapcache page, and then reuse its swap for
666 * another page of the image. On waking from hibernation, the
667 * original page might be freed under memory pressure, then
668 * later read back in from swap, now with the wrong data.
670 * Hibration suspends storage while it is writing the image
671 * to disk so check that here.
673 if (pm_suspended_storage())
676 delete_from_swap_cache(page
);
682 * Free the swap entry like above, but also try to
683 * free the page cache entry if it is the last user.
685 int free_swap_and_cache(swp_entry_t entry
)
687 struct swap_info_struct
*p
;
688 struct page
*page
= NULL
;
690 if (non_swap_entry(entry
))
693 p
= swap_info_get(entry
);
695 if (swap_entry_free(p
, entry
, 1) == SWAP_HAS_CACHE
) {
696 page
= find_get_page(&swapper_space
, entry
.val
);
697 if (page
&& !trylock_page(page
)) {
698 page_cache_release(page
);
702 spin_unlock(&swap_lock
);
706 * Not mapped elsewhere, or swap space full? Free it!
707 * Also recheck PageSwapCache now page is locked (above).
709 if (PageSwapCache(page
) && !PageWriteback(page
) &&
710 (!page_mapped(page
) || vm_swap_full())) {
711 delete_from_swap_cache(page
);
715 page_cache_release(page
);
720 #ifdef CONFIG_HIBERNATION
722 * Find the swap type that corresponds to given device (if any).
724 * @offset - number of the PAGE_SIZE-sized block of the device, starting
725 * from 0, in which the swap header is expected to be located.
727 * This is needed for the suspend to disk (aka swsusp).
729 int swap_type_of(dev_t device
, sector_t offset
, struct block_device
**bdev_p
)
731 struct block_device
*bdev
= NULL
;
735 bdev
= bdget(device
);
737 spin_lock(&swap_lock
);
738 for (type
= 0; type
< nr_swapfiles
; type
++) {
739 struct swap_info_struct
*sis
= swap_info
[type
];
741 if (!(sis
->flags
& SWP_WRITEOK
))
746 *bdev_p
= bdgrab(sis
->bdev
);
748 spin_unlock(&swap_lock
);
751 if (bdev
== sis
->bdev
) {
752 struct swap_extent
*se
= &sis
->first_swap_extent
;
754 if (se
->start_block
== offset
) {
756 *bdev_p
= bdgrab(sis
->bdev
);
758 spin_unlock(&swap_lock
);
764 spin_unlock(&swap_lock
);
772 * Get the (PAGE_SIZE) block corresponding to given offset on the swapdev
773 * corresponding to given index in swap_info (swap type).
775 sector_t
swapdev_block(int type
, pgoff_t offset
)
777 struct block_device
*bdev
;
779 if ((unsigned int)type
>= nr_swapfiles
)
781 if (!(swap_info
[type
]->flags
& SWP_WRITEOK
))
783 return map_swap_entry(swp_entry(type
, offset
), &bdev
);
787 * Return either the total number of swap pages of given type, or the number
788 * of free pages of that type (depending on @free)
790 * This is needed for software suspend
792 unsigned int count_swap_pages(int type
, int free
)
796 spin_lock(&swap_lock
);
797 if ((unsigned int)type
< nr_swapfiles
) {
798 struct swap_info_struct
*sis
= swap_info
[type
];
800 if (sis
->flags
& SWP_WRITEOK
) {
803 n
-= sis
->inuse_pages
;
806 spin_unlock(&swap_lock
);
809 #endif /* CONFIG_HIBERNATION */
812 * No need to decide whether this PTE shares the swap entry with others,
813 * just let do_wp_page work it out if a write is requested later - to
814 * force COW, vm_page_prot omits write permission from any private vma.
816 static int unuse_pte(struct vm_area_struct
*vma
, pmd_t
*pmd
,
817 unsigned long addr
, swp_entry_t entry
, struct page
*page
)
819 struct mem_cgroup
*memcg
;
824 if (mem_cgroup_try_charge_swapin(vma
->vm_mm
, page
,
825 GFP_KERNEL
, &memcg
)) {
830 pte
= pte_offset_map_lock(vma
->vm_mm
, pmd
, addr
, &ptl
);
831 if (unlikely(!pte_same(*pte
, swp_entry_to_pte(entry
)))) {
833 mem_cgroup_cancel_charge_swapin(memcg
);
838 dec_mm_counter(vma
->vm_mm
, MM_SWAPENTS
);
839 inc_mm_counter(vma
->vm_mm
, MM_ANONPAGES
);
841 set_pte_at(vma
->vm_mm
, addr
, pte
,
842 pte_mkold(mk_pte(page
, vma
->vm_page_prot
)));
843 page_add_anon_rmap(page
, vma
, addr
);
844 mem_cgroup_commit_charge_swapin(page
, memcg
);
847 * Move the page to the active list so it is not
848 * immediately swapped out again after swapon.
852 pte_unmap_unlock(pte
, ptl
);
857 static int unuse_pte_range(struct vm_area_struct
*vma
, pmd_t
*pmd
,
858 unsigned long addr
, unsigned long end
,
859 swp_entry_t entry
, struct page
*page
)
861 pte_t swp_pte
= swp_entry_to_pte(entry
);
866 * We don't actually need pte lock while scanning for swp_pte: since
867 * we hold page lock and mmap_sem, swp_pte cannot be inserted into the
868 * page table while we're scanning; though it could get zapped, and on
869 * some architectures (e.g. x86_32 with PAE) we might catch a glimpse
870 * of unmatched parts which look like swp_pte, so unuse_pte must
871 * recheck under pte lock. Scanning without pte lock lets it be
872 * preemptible whenever CONFIG_PREEMPT but not CONFIG_HIGHPTE.
874 pte
= pte_offset_map(pmd
, addr
);
877 * swapoff spends a _lot_ of time in this loop!
878 * Test inline before going to call unuse_pte.
880 if (unlikely(pte_same(*pte
, swp_pte
))) {
882 ret
= unuse_pte(vma
, pmd
, addr
, entry
, page
);
885 pte
= pte_offset_map(pmd
, addr
);
887 } while (pte
++, addr
+= PAGE_SIZE
, addr
!= end
);
893 static inline int unuse_pmd_range(struct vm_area_struct
*vma
, pud_t
*pud
,
894 unsigned long addr
, unsigned long end
,
895 swp_entry_t entry
, struct page
*page
)
901 pmd
= pmd_offset(pud
, addr
);
903 next
= pmd_addr_end(addr
, end
);
904 if (pmd_none_or_trans_huge_or_clear_bad(pmd
))
906 ret
= unuse_pte_range(vma
, pmd
, addr
, next
, entry
, page
);
909 } while (pmd
++, addr
= next
, addr
!= end
);
913 static inline int unuse_pud_range(struct vm_area_struct
*vma
, pgd_t
*pgd
,
914 unsigned long addr
, unsigned long end
,
915 swp_entry_t entry
, struct page
*page
)
921 pud
= pud_offset(pgd
, addr
);
923 next
= pud_addr_end(addr
, end
);
924 if (pud_none_or_clear_bad(pud
))
926 ret
= unuse_pmd_range(vma
, pud
, addr
, next
, entry
, page
);
929 } while (pud
++, addr
= next
, addr
!= end
);
933 static int unuse_vma(struct vm_area_struct
*vma
,
934 swp_entry_t entry
, struct page
*page
)
937 unsigned long addr
, end
, next
;
940 if (page_anon_vma(page
)) {
941 addr
= page_address_in_vma(page
, vma
);
945 end
= addr
+ PAGE_SIZE
;
947 addr
= vma
->vm_start
;
951 pgd
= pgd_offset(vma
->vm_mm
, addr
);
953 next
= pgd_addr_end(addr
, end
);
954 if (pgd_none_or_clear_bad(pgd
))
956 ret
= unuse_pud_range(vma
, pgd
, addr
, next
, entry
, page
);
959 } while (pgd
++, addr
= next
, addr
!= end
);
963 static int unuse_mm(struct mm_struct
*mm
,
964 swp_entry_t entry
, struct page
*page
)
966 struct vm_area_struct
*vma
;
969 if (!down_read_trylock(&mm
->mmap_sem
)) {
971 * Activate page so shrink_inactive_list is unlikely to unmap
972 * its ptes while lock is dropped, so swapoff can make progress.
976 down_read(&mm
->mmap_sem
);
979 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
980 if (vma
->anon_vma
&& (ret
= unuse_vma(vma
, entry
, page
)))
983 up_read(&mm
->mmap_sem
);
984 return (ret
< 0)? ret
: 0;
988 * Scan swap_map from current position to next entry still in use.
989 * Recycle to start on reaching the end, returning 0 when empty.
991 static unsigned int find_next_to_unuse(struct swap_info_struct
*si
,
994 unsigned int max
= si
->max
;
995 unsigned int i
= prev
;
999 * No need for swap_lock here: we're just looking
1000 * for whether an entry is in use, not modifying it; false
1001 * hits are okay, and sys_swapoff() has already prevented new
1002 * allocations from this area (while holding swap_lock).
1011 * No entries in use at top of swap_map,
1012 * loop back to start and recheck there.
1018 count
= si
->swap_map
[i
];
1019 if (count
&& swap_count(count
) != SWAP_MAP_BAD
)
1026 * We completely avoid races by reading each swap page in advance,
1027 * and then search for the process using it. All the necessary
1028 * page table adjustments can then be made atomically.
1030 static int try_to_unuse(unsigned int type
)
1032 struct swap_info_struct
*si
= swap_info
[type
];
1033 struct mm_struct
*start_mm
;
1034 unsigned char *swap_map
;
1035 unsigned char swcount
;
1042 * When searching mms for an entry, a good strategy is to
1043 * start at the first mm we freed the previous entry from
1044 * (though actually we don't notice whether we or coincidence
1045 * freed the entry). Initialize this start_mm with a hold.
1047 * A simpler strategy would be to start at the last mm we
1048 * freed the previous entry from; but that would take less
1049 * advantage of mmlist ordering, which clusters forked mms
1050 * together, child after parent. If we race with dup_mmap(), we
1051 * prefer to resolve parent before child, lest we miss entries
1052 * duplicated after we scanned child: using last mm would invert
1055 start_mm
= &init_mm
;
1056 atomic_inc(&init_mm
.mm_users
);
1059 * Keep on scanning until all entries have gone. Usually,
1060 * one pass through swap_map is enough, but not necessarily:
1061 * there are races when an instance of an entry might be missed.
1063 while ((i
= find_next_to_unuse(si
, i
)) != 0) {
1064 if (signal_pending(current
)) {
1070 * Get a page for the entry, using the existing swap
1071 * cache page if there is one. Otherwise, get a clean
1072 * page and read the swap into it.
1074 swap_map
= &si
->swap_map
[i
];
1075 entry
= swp_entry(type
, i
);
1076 page
= read_swap_cache_async(entry
,
1077 GFP_HIGHUSER_MOVABLE
, NULL
, 0);
1080 * Either swap_duplicate() failed because entry
1081 * has been freed independently, and will not be
1082 * reused since sys_swapoff() already disabled
1083 * allocation from here, or alloc_page() failed.
1092 * Don't hold on to start_mm if it looks like exiting.
1094 if (atomic_read(&start_mm
->mm_users
) == 1) {
1096 start_mm
= &init_mm
;
1097 atomic_inc(&init_mm
.mm_users
);
1101 * Wait for and lock page. When do_swap_page races with
1102 * try_to_unuse, do_swap_page can handle the fault much
1103 * faster than try_to_unuse can locate the entry. This
1104 * apparently redundant "wait_on_page_locked" lets try_to_unuse
1105 * defer to do_swap_page in such a case - in some tests,
1106 * do_swap_page and try_to_unuse repeatedly compete.
1108 wait_on_page_locked(page
);
1109 wait_on_page_writeback(page
);
1111 wait_on_page_writeback(page
);
1114 * Remove all references to entry.
1116 swcount
= *swap_map
;
1117 if (swap_count(swcount
) == SWAP_MAP_SHMEM
) {
1118 retval
= shmem_unuse(entry
, page
);
1119 /* page has already been unlocked and released */
1124 if (swap_count(swcount
) && start_mm
!= &init_mm
)
1125 retval
= unuse_mm(start_mm
, entry
, page
);
1127 if (swap_count(*swap_map
)) {
1128 int set_start_mm
= (*swap_map
>= swcount
);
1129 struct list_head
*p
= &start_mm
->mmlist
;
1130 struct mm_struct
*new_start_mm
= start_mm
;
1131 struct mm_struct
*prev_mm
= start_mm
;
1132 struct mm_struct
*mm
;
1134 atomic_inc(&new_start_mm
->mm_users
);
1135 atomic_inc(&prev_mm
->mm_users
);
1136 spin_lock(&mmlist_lock
);
1137 while (swap_count(*swap_map
) && !retval
&&
1138 (p
= p
->next
) != &start_mm
->mmlist
) {
1139 mm
= list_entry(p
, struct mm_struct
, mmlist
);
1140 if (!atomic_inc_not_zero(&mm
->mm_users
))
1142 spin_unlock(&mmlist_lock
);
1148 swcount
= *swap_map
;
1149 if (!swap_count(swcount
)) /* any usage ? */
1151 else if (mm
== &init_mm
)
1154 retval
= unuse_mm(mm
, entry
, page
);
1156 if (set_start_mm
&& *swap_map
< swcount
) {
1157 mmput(new_start_mm
);
1158 atomic_inc(&mm
->mm_users
);
1162 spin_lock(&mmlist_lock
);
1164 spin_unlock(&mmlist_lock
);
1167 start_mm
= new_start_mm
;
1171 page_cache_release(page
);
1176 * If a reference remains (rare), we would like to leave
1177 * the page in the swap cache; but try_to_unmap could
1178 * then re-duplicate the entry once we drop page lock,
1179 * so we might loop indefinitely; also, that page could
1180 * not be swapped out to other storage meanwhile. So:
1181 * delete from cache even if there's another reference,
1182 * after ensuring that the data has been saved to disk -
1183 * since if the reference remains (rarer), it will be
1184 * read from disk into another page. Splitting into two
1185 * pages would be incorrect if swap supported "shared
1186 * private" pages, but they are handled by tmpfs files.
1188 * Given how unuse_vma() targets one particular offset
1189 * in an anon_vma, once the anon_vma has been determined,
1190 * this splitting happens to be just what is needed to
1191 * handle where KSM pages have been swapped out: re-reading
1192 * is unnecessarily slow, but we can fix that later on.
1194 if (swap_count(*swap_map
) &&
1195 PageDirty(page
) && PageSwapCache(page
)) {
1196 struct writeback_control wbc
= {
1197 .sync_mode
= WB_SYNC_NONE
,
1200 swap_writepage(page
, &wbc
);
1202 wait_on_page_writeback(page
);
1206 * It is conceivable that a racing task removed this page from
1207 * swap cache just before we acquired the page lock at the top,
1208 * or while we dropped it in unuse_mm(). The page might even
1209 * be back in swap cache on another swap area: that we must not
1210 * delete, since it may not have been written out to swap yet.
1212 if (PageSwapCache(page
) &&
1213 likely(page_private(page
) == entry
.val
))
1214 delete_from_swap_cache(page
);
1217 * So we could skip searching mms once swap count went
1218 * to 1, we did not mark any present ptes as dirty: must
1219 * mark page dirty so shrink_page_list will preserve it.
1223 page_cache_release(page
);
1226 * Make sure that we aren't completely killing
1227 * interactive performance.
1237 * After a successful try_to_unuse, if no swap is now in use, we know
1238 * we can empty the mmlist. swap_lock must be held on entry and exit.
1239 * Note that mmlist_lock nests inside swap_lock, and an mm must be
1240 * added to the mmlist just after page_duplicate - before would be racy.
1242 static void drain_mmlist(void)
1244 struct list_head
*p
, *next
;
1247 for (type
= 0; type
< nr_swapfiles
; type
++)
1248 if (swap_info
[type
]->inuse_pages
)
1250 spin_lock(&mmlist_lock
);
1251 list_for_each_safe(p
, next
, &init_mm
.mmlist
)
1253 spin_unlock(&mmlist_lock
);
1257 * Use this swapdev's extent info to locate the (PAGE_SIZE) block which
1258 * corresponds to page offset for the specified swap entry.
1259 * Note that the type of this function is sector_t, but it returns page offset
1260 * into the bdev, not sector offset.
1262 static sector_t
map_swap_entry(swp_entry_t entry
, struct block_device
**bdev
)
1264 struct swap_info_struct
*sis
;
1265 struct swap_extent
*start_se
;
1266 struct swap_extent
*se
;
1269 sis
= swap_info
[swp_type(entry
)];
1272 offset
= swp_offset(entry
);
1273 start_se
= sis
->curr_swap_extent
;
1277 struct list_head
*lh
;
1279 if (se
->start_page
<= offset
&&
1280 offset
< (se
->start_page
+ se
->nr_pages
)) {
1281 return se
->start_block
+ (offset
- se
->start_page
);
1284 se
= list_entry(lh
, struct swap_extent
, list
);
1285 sis
->curr_swap_extent
= se
;
1286 BUG_ON(se
== start_se
); /* It *must* be present */
1291 * Returns the page offset into bdev for the specified page's swap entry.
1293 sector_t
map_swap_page(struct page
*page
, struct block_device
**bdev
)
1296 entry
.val
= page_private(page
);
1297 return map_swap_entry(entry
, bdev
);
1301 * Free all of a swapdev's extent information
1303 static void destroy_swap_extents(struct swap_info_struct
*sis
)
1305 while (!list_empty(&sis
->first_swap_extent
.list
)) {
1306 struct swap_extent
*se
;
1308 se
= list_entry(sis
->first_swap_extent
.list
.next
,
1309 struct swap_extent
, list
);
1310 list_del(&se
->list
);
1316 * Add a block range (and the corresponding page range) into this swapdev's
1317 * extent list. The extent list is kept sorted in page order.
1319 * This function rather assumes that it is called in ascending page order.
1322 add_swap_extent(struct swap_info_struct
*sis
, unsigned long start_page
,
1323 unsigned long nr_pages
, sector_t start_block
)
1325 struct swap_extent
*se
;
1326 struct swap_extent
*new_se
;
1327 struct list_head
*lh
;
1329 if (start_page
== 0) {
1330 se
= &sis
->first_swap_extent
;
1331 sis
->curr_swap_extent
= se
;
1333 se
->nr_pages
= nr_pages
;
1334 se
->start_block
= start_block
;
1337 lh
= sis
->first_swap_extent
.list
.prev
; /* Highest extent */
1338 se
= list_entry(lh
, struct swap_extent
, list
);
1339 BUG_ON(se
->start_page
+ se
->nr_pages
!= start_page
);
1340 if (se
->start_block
+ se
->nr_pages
== start_block
) {
1342 se
->nr_pages
+= nr_pages
;
1348 * No merge. Insert a new extent, preserving ordering.
1350 new_se
= kmalloc(sizeof(*se
), GFP_KERNEL
);
1353 new_se
->start_page
= start_page
;
1354 new_se
->nr_pages
= nr_pages
;
1355 new_se
->start_block
= start_block
;
1357 list_add_tail(&new_se
->list
, &sis
->first_swap_extent
.list
);
1362 * A `swap extent' is a simple thing which maps a contiguous range of pages
1363 * onto a contiguous range of disk blocks. An ordered list of swap extents
1364 * is built at swapon time and is then used at swap_writepage/swap_readpage
1365 * time for locating where on disk a page belongs.
1367 * If the swapfile is an S_ISBLK block device, a single extent is installed.
1368 * This is done so that the main operating code can treat S_ISBLK and S_ISREG
1369 * swap files identically.
1371 * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap
1372 * extent list operates in PAGE_SIZE disk blocks. Both S_ISREG and S_ISBLK
1373 * swapfiles are handled *identically* after swapon time.
1375 * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks
1376 * and will parse them into an ordered extent list, in PAGE_SIZE chunks. If
1377 * some stray blocks are found which do not fall within the PAGE_SIZE alignment
1378 * requirements, they are simply tossed out - we will never use those blocks
1381 * For S_ISREG swapfiles we set S_SWAPFILE across the life of the swapon. This
1382 * prevents root from shooting her foot off by ftruncating an in-use swapfile,
1383 * which will scribble on the fs.
1385 * The amount of disk space which a single swap extent represents varies.
1386 * Typically it is in the 1-4 megabyte range. So we can have hundreds of
1387 * extents in the list. To avoid much list walking, we cache the previous
1388 * search location in `curr_swap_extent', and start new searches from there.
1389 * This is extremely effective. The average number of iterations in
1390 * map_swap_page() has been measured at about 0.3 per page. - akpm.
1392 static int setup_swap_extents(struct swap_info_struct
*sis
, sector_t
*span
)
1394 struct inode
*inode
;
1395 unsigned blocks_per_page
;
1396 unsigned long page_no
;
1398 sector_t probe_block
;
1399 sector_t last_block
;
1400 sector_t lowest_block
= -1;
1401 sector_t highest_block
= 0;
1405 inode
= sis
->swap_file
->f_mapping
->host
;
1406 if (S_ISBLK(inode
->i_mode
)) {
1407 ret
= add_swap_extent(sis
, 0, sis
->max
, 0);
1412 blkbits
= inode
->i_blkbits
;
1413 blocks_per_page
= PAGE_SIZE
>> blkbits
;
1416 * Map all the blocks into the extent list. This code doesn't try
1421 last_block
= i_size_read(inode
) >> blkbits
;
1422 while ((probe_block
+ blocks_per_page
) <= last_block
&&
1423 page_no
< sis
->max
) {
1424 unsigned block_in_page
;
1425 sector_t first_block
;
1427 first_block
= bmap(inode
, probe_block
);
1428 if (first_block
== 0)
1432 * It must be PAGE_SIZE aligned on-disk
1434 if (first_block
& (blocks_per_page
- 1)) {
1439 for (block_in_page
= 1; block_in_page
< blocks_per_page
;
1443 block
= bmap(inode
, probe_block
+ block_in_page
);
1446 if (block
!= first_block
+ block_in_page
) {
1453 first_block
>>= (PAGE_SHIFT
- blkbits
);
1454 if (page_no
) { /* exclude the header page */
1455 if (first_block
< lowest_block
)
1456 lowest_block
= first_block
;
1457 if (first_block
> highest_block
)
1458 highest_block
= first_block
;
1462 * We found a PAGE_SIZE-length, PAGE_SIZE-aligned run of blocks
1464 ret
= add_swap_extent(sis
, page_no
, 1, first_block
);
1469 probe_block
+= blocks_per_page
;
1474 *span
= 1 + highest_block
- lowest_block
;
1476 page_no
= 1; /* force Empty message */
1478 sis
->pages
= page_no
- 1;
1479 sis
->highest_bit
= page_no
- 1;
1483 printk(KERN_ERR
"swapon: swapfile has holes\n");
1488 static void enable_swap_info(struct swap_info_struct
*p
, int prio
,
1489 unsigned char *swap_map
)
1493 spin_lock(&swap_lock
);
1497 p
->prio
= --least_priority
;
1498 p
->swap_map
= swap_map
;
1499 p
->flags
|= SWP_WRITEOK
;
1500 nr_swap_pages
+= p
->pages
;
1501 total_swap_pages
+= p
->pages
;
1503 /* insert swap space into swap_list: */
1505 for (i
= swap_list
.head
; i
>= 0; i
= swap_info
[i
]->next
) {
1506 if (p
->prio
>= swap_info
[i
]->prio
)
1512 swap_list
.head
= swap_list
.next
= p
->type
;
1514 swap_info
[prev
]->next
= p
->type
;
1515 spin_unlock(&swap_lock
);
1518 SYSCALL_DEFINE1(swapoff
, const char __user
*, specialfile
)
1520 struct swap_info_struct
*p
= NULL
;
1521 unsigned char *swap_map
;
1522 struct file
*swap_file
, *victim
;
1523 struct address_space
*mapping
;
1524 struct inode
*inode
;
1530 if (!capable(CAP_SYS_ADMIN
))
1533 BUG_ON(!current
->mm
);
1535 pathname
= getname(specialfile
);
1536 err
= PTR_ERR(pathname
);
1537 if (IS_ERR(pathname
))
1540 victim
= filp_open(pathname
, O_RDWR
|O_LARGEFILE
, 0);
1542 err
= PTR_ERR(victim
);
1546 mapping
= victim
->f_mapping
;
1548 spin_lock(&swap_lock
);
1549 for (type
= swap_list
.head
; type
>= 0; type
= swap_info
[type
]->next
) {
1550 p
= swap_info
[type
];
1551 if (p
->flags
& SWP_WRITEOK
) {
1552 if (p
->swap_file
->f_mapping
== mapping
)
1559 spin_unlock(&swap_lock
);
1562 if (!security_vm_enough_memory_mm(current
->mm
, p
->pages
))
1563 vm_unacct_memory(p
->pages
);
1566 spin_unlock(&swap_lock
);
1570 swap_list
.head
= p
->next
;
1572 swap_info
[prev
]->next
= p
->next
;
1573 if (type
== swap_list
.next
) {
1574 /* just pick something that's safe... */
1575 swap_list
.next
= swap_list
.head
;
1578 for (i
= p
->next
; i
>= 0; i
= swap_info
[i
]->next
)
1579 swap_info
[i
]->prio
= p
->prio
--;
1582 nr_swap_pages
-= p
->pages
;
1583 total_swap_pages
-= p
->pages
;
1584 p
->flags
&= ~SWP_WRITEOK
;
1585 spin_unlock(&swap_lock
);
1587 oom_score_adj
= test_set_oom_score_adj(OOM_SCORE_ADJ_MAX
);
1588 err
= try_to_unuse(type
);
1589 compare_swap_oom_score_adj(OOM_SCORE_ADJ_MAX
, oom_score_adj
);
1593 * reading p->prio and p->swap_map outside the lock is
1594 * safe here because only sys_swapon and sys_swapoff
1595 * change them, and there can be no other sys_swapon or
1596 * sys_swapoff for this swap_info_struct at this point.
1598 /* re-insert swap space back into swap_list */
1599 enable_swap_info(p
, p
->prio
, p
->swap_map
);
1603 destroy_swap_extents(p
);
1604 if (p
->flags
& SWP_CONTINUED
)
1605 free_swap_count_continuations(p
);
1607 mutex_lock(&swapon_mutex
);
1608 spin_lock(&swap_lock
);
1611 /* wait for anyone still in scan_swap_map */
1612 p
->highest_bit
= 0; /* cuts scans short */
1613 while (p
->flags
>= SWP_SCANNING
) {
1614 spin_unlock(&swap_lock
);
1615 schedule_timeout_uninterruptible(1);
1616 spin_lock(&swap_lock
);
1619 swap_file
= p
->swap_file
;
1620 p
->swap_file
= NULL
;
1622 swap_map
= p
->swap_map
;
1625 spin_unlock(&swap_lock
);
1626 mutex_unlock(&swapon_mutex
);
1628 /* Destroy swap account informatin */
1629 swap_cgroup_swapoff(type
);
1631 inode
= mapping
->host
;
1632 if (S_ISBLK(inode
->i_mode
)) {
1633 struct block_device
*bdev
= I_BDEV(inode
);
1634 set_blocksize(bdev
, p
->old_block_size
);
1635 blkdev_put(bdev
, FMODE_READ
| FMODE_WRITE
| FMODE_EXCL
);
1637 mutex_lock(&inode
->i_mutex
);
1638 inode
->i_flags
&= ~S_SWAPFILE
;
1639 mutex_unlock(&inode
->i_mutex
);
1641 filp_close(swap_file
, NULL
);
1643 atomic_inc(&proc_poll_event
);
1644 wake_up_interruptible(&proc_poll_wait
);
1647 filp_close(victim
, NULL
);
1652 #ifdef CONFIG_PROC_FS
1653 static unsigned swaps_poll(struct file
*file
, poll_table
*wait
)
1655 struct seq_file
*seq
= file
->private_data
;
1657 poll_wait(file
, &proc_poll_wait
, wait
);
1659 if (seq
->poll_event
!= atomic_read(&proc_poll_event
)) {
1660 seq
->poll_event
= atomic_read(&proc_poll_event
);
1661 return POLLIN
| POLLRDNORM
| POLLERR
| POLLPRI
;
1664 return POLLIN
| POLLRDNORM
;
1668 static void *swap_start(struct seq_file
*swap
, loff_t
*pos
)
1670 struct swap_info_struct
*si
;
1674 mutex_lock(&swapon_mutex
);
1677 return SEQ_START_TOKEN
;
1679 for (type
= 0; type
< nr_swapfiles
; type
++) {
1680 smp_rmb(); /* read nr_swapfiles before swap_info[type] */
1681 si
= swap_info
[type
];
1682 if (!(si
->flags
& SWP_USED
) || !si
->swap_map
)
1691 static void *swap_next(struct seq_file
*swap
, void *v
, loff_t
*pos
)
1693 struct swap_info_struct
*si
= v
;
1696 if (v
== SEQ_START_TOKEN
)
1699 type
= si
->type
+ 1;
1701 for (; type
< nr_swapfiles
; type
++) {
1702 smp_rmb(); /* read nr_swapfiles before swap_info[type] */
1703 si
= swap_info
[type
];
1704 if (!(si
->flags
& SWP_USED
) || !si
->swap_map
)
1713 static void swap_stop(struct seq_file
*swap
, void *v
)
1715 mutex_unlock(&swapon_mutex
);
1718 static int swap_show(struct seq_file
*swap
, void *v
)
1720 struct swap_info_struct
*si
= v
;
1724 if (si
== SEQ_START_TOKEN
) {
1725 seq_puts(swap
,"Filename\t\t\t\tType\t\tSize\tUsed\tPriority\n");
1729 file
= si
->swap_file
;
1730 len
= seq_path(swap
, &file
->f_path
, " \t\n\\");
1731 seq_printf(swap
, "%*s%s\t%u\t%u\t%d\n",
1732 len
< 40 ? 40 - len
: 1, " ",
1733 S_ISBLK(file
->f_path
.dentry
->d_inode
->i_mode
) ?
1734 "partition" : "file\t",
1735 si
->pages
<< (PAGE_SHIFT
- 10),
1736 si
->inuse_pages
<< (PAGE_SHIFT
- 10),
1741 static const struct seq_operations swaps_op
= {
1742 .start
= swap_start
,
1748 static int swaps_open(struct inode
*inode
, struct file
*file
)
1750 struct seq_file
*seq
;
1753 ret
= seq_open(file
, &swaps_op
);
1757 seq
= file
->private_data
;
1758 seq
->poll_event
= atomic_read(&proc_poll_event
);
1762 static const struct file_operations proc_swaps_operations
= {
1765 .llseek
= seq_lseek
,
1766 .release
= seq_release
,
1770 static int __init
procswaps_init(void)
1772 proc_create("swaps", 0, NULL
, &proc_swaps_operations
);
1775 __initcall(procswaps_init
);
1776 #endif /* CONFIG_PROC_FS */
1778 #ifdef MAX_SWAPFILES_CHECK
1779 static int __init
max_swapfiles_check(void)
1781 MAX_SWAPFILES_CHECK();
1784 late_initcall(max_swapfiles_check
);
1787 static struct swap_info_struct
*alloc_swap_info(void)
1789 struct swap_info_struct
*p
;
1792 p
= kzalloc(sizeof(*p
), GFP_KERNEL
);
1794 return ERR_PTR(-ENOMEM
);
1796 spin_lock(&swap_lock
);
1797 for (type
= 0; type
< nr_swapfiles
; type
++) {
1798 if (!(swap_info
[type
]->flags
& SWP_USED
))
1801 if (type
>= MAX_SWAPFILES
) {
1802 spin_unlock(&swap_lock
);
1804 return ERR_PTR(-EPERM
);
1806 if (type
>= nr_swapfiles
) {
1808 swap_info
[type
] = p
;
1810 * Write swap_info[type] before nr_swapfiles, in case a
1811 * racing procfs swap_start() or swap_next() is reading them.
1812 * (We never shrink nr_swapfiles, we never free this entry.)
1818 p
= swap_info
[type
];
1820 * Do not memset this entry: a racing procfs swap_next()
1821 * would be relying on p->type to remain valid.
1824 INIT_LIST_HEAD(&p
->first_swap_extent
.list
);
1825 p
->flags
= SWP_USED
;
1827 spin_unlock(&swap_lock
);
1832 static int claim_swapfile(struct swap_info_struct
*p
, struct inode
*inode
)
1836 if (S_ISBLK(inode
->i_mode
)) {
1837 p
->bdev
= bdgrab(I_BDEV(inode
));
1838 error
= blkdev_get(p
->bdev
,
1839 FMODE_READ
| FMODE_WRITE
| FMODE_EXCL
,
1845 p
->old_block_size
= block_size(p
->bdev
);
1846 error
= set_blocksize(p
->bdev
, PAGE_SIZE
);
1849 p
->flags
|= SWP_BLKDEV
;
1850 } else if (S_ISREG(inode
->i_mode
)) {
1851 p
->bdev
= inode
->i_sb
->s_bdev
;
1852 mutex_lock(&inode
->i_mutex
);
1853 if (IS_SWAPFILE(inode
))
1861 static unsigned long read_swap_header(struct swap_info_struct
*p
,
1862 union swap_header
*swap_header
,
1863 struct inode
*inode
)
1866 unsigned long maxpages
;
1867 unsigned long swapfilepages
;
1869 if (memcmp("SWAPSPACE2", swap_header
->magic
.magic
, 10)) {
1870 printk(KERN_ERR
"Unable to find swap-space signature\n");
1874 /* swap partition endianess hack... */
1875 if (swab32(swap_header
->info
.version
) == 1) {
1876 swab32s(&swap_header
->info
.version
);
1877 swab32s(&swap_header
->info
.last_page
);
1878 swab32s(&swap_header
->info
.nr_badpages
);
1879 for (i
= 0; i
< swap_header
->info
.nr_badpages
; i
++)
1880 swab32s(&swap_header
->info
.badpages
[i
]);
1882 /* Check the swap header's sub-version */
1883 if (swap_header
->info
.version
!= 1) {
1885 "Unable to handle swap header version %d\n",
1886 swap_header
->info
.version
);
1891 p
->cluster_next
= 1;
1895 * Find out how many pages are allowed for a single swap
1896 * device. There are three limiting factors: 1) the number
1897 * of bits for the swap offset in the swp_entry_t type, and
1898 * 2) the number of bits in the swap pte as defined by the
1899 * the different architectures, and 3) the number of free bits
1900 * in an exceptional radix_tree entry. In order to find the
1901 * largest possible bit mask, a swap entry with swap type 0
1902 * and swap offset ~0UL is created, encoded to a swap pte,
1903 * decoded to a swp_entry_t again, and finally the swap
1904 * offset is extracted. This will mask all the bits from
1905 * the initial ~0UL mask that can't be encoded in either
1906 * the swp_entry_t or the architecture definition of a
1907 * swap pte. Then the same is done for a radix_tree entry.
1909 maxpages
= swp_offset(pte_to_swp_entry(
1910 swp_entry_to_pte(swp_entry(0, ~0UL))));
1911 maxpages
= swp_offset(radix_to_swp_entry(
1912 swp_to_radix_entry(swp_entry(0, maxpages
)))) + 1;
1914 if (maxpages
> swap_header
->info
.last_page
) {
1915 maxpages
= swap_header
->info
.last_page
+ 1;
1916 /* p->max is an unsigned int: don't overflow it */
1917 if ((unsigned int)maxpages
== 0)
1918 maxpages
= UINT_MAX
;
1920 p
->highest_bit
= maxpages
- 1;
1924 swapfilepages
= i_size_read(inode
) >> PAGE_SHIFT
;
1925 if (swapfilepages
&& maxpages
> swapfilepages
) {
1927 "Swap area shorter than signature indicates\n");
1930 if (swap_header
->info
.nr_badpages
&& S_ISREG(inode
->i_mode
))
1932 if (swap_header
->info
.nr_badpages
> MAX_SWAP_BADPAGES
)
1938 static int setup_swap_map_and_extents(struct swap_info_struct
*p
,
1939 union swap_header
*swap_header
,
1940 unsigned char *swap_map
,
1941 unsigned long maxpages
,
1945 unsigned int nr_good_pages
;
1948 nr_good_pages
= maxpages
- 1; /* omit header page */
1950 for (i
= 0; i
< swap_header
->info
.nr_badpages
; i
++) {
1951 unsigned int page_nr
= swap_header
->info
.badpages
[i
];
1952 if (page_nr
== 0 || page_nr
> swap_header
->info
.last_page
)
1954 if (page_nr
< maxpages
) {
1955 swap_map
[page_nr
] = SWAP_MAP_BAD
;
1960 if (nr_good_pages
) {
1961 swap_map
[0] = SWAP_MAP_BAD
;
1963 p
->pages
= nr_good_pages
;
1964 nr_extents
= setup_swap_extents(p
, span
);
1967 nr_good_pages
= p
->pages
;
1969 if (!nr_good_pages
) {
1970 printk(KERN_WARNING
"Empty swap-file\n");
1977 SYSCALL_DEFINE2(swapon
, const char __user
*, specialfile
, int, swap_flags
)
1979 struct swap_info_struct
*p
;
1981 struct file
*swap_file
= NULL
;
1982 struct address_space
*mapping
;
1986 union swap_header
*swap_header
;
1989 unsigned long maxpages
;
1990 unsigned char *swap_map
= NULL
;
1991 struct page
*page
= NULL
;
1992 struct inode
*inode
= NULL
;
1994 if (swap_flags
& ~SWAP_FLAGS_VALID
)
1997 if (!capable(CAP_SYS_ADMIN
))
2000 p
= alloc_swap_info();
2004 name
= getname(specialfile
);
2006 error
= PTR_ERR(name
);
2010 swap_file
= filp_open(name
, O_RDWR
|O_LARGEFILE
, 0);
2011 if (IS_ERR(swap_file
)) {
2012 error
= PTR_ERR(swap_file
);
2017 p
->swap_file
= swap_file
;
2018 mapping
= swap_file
->f_mapping
;
2020 for (i
= 0; i
< nr_swapfiles
; i
++) {
2021 struct swap_info_struct
*q
= swap_info
[i
];
2023 if (q
== p
|| !q
->swap_file
)
2025 if (mapping
== q
->swap_file
->f_mapping
) {
2031 inode
= mapping
->host
;
2032 /* If S_ISREG(inode->i_mode) will do mutex_lock(&inode->i_mutex); */
2033 error
= claim_swapfile(p
, inode
);
2034 if (unlikely(error
))
2038 * Read the swap header.
2040 if (!mapping
->a_ops
->readpage
) {
2044 page
= read_mapping_page(mapping
, 0, swap_file
);
2046 error
= PTR_ERR(page
);
2049 swap_header
= kmap(page
);
2051 maxpages
= read_swap_header(p
, swap_header
, inode
);
2052 if (unlikely(!maxpages
)) {
2057 /* OK, set up the swap map and apply the bad block list */
2058 swap_map
= vzalloc(maxpages
);
2064 error
= swap_cgroup_swapon(p
->type
, maxpages
);
2068 nr_extents
= setup_swap_map_and_extents(p
, swap_header
, swap_map
,
2070 if (unlikely(nr_extents
< 0)) {
2076 if (blk_queue_nonrot(bdev_get_queue(p
->bdev
))) {
2077 p
->flags
|= SWP_SOLIDSTATE
;
2078 p
->cluster_next
= 1 + (random32() % p
->highest_bit
);
2080 if ((swap_flags
& SWAP_FLAG_DISCARD
) && discard_swap(p
) == 0)
2081 p
->flags
|= SWP_DISCARDABLE
;
2084 mutex_lock(&swapon_mutex
);
2086 if (swap_flags
& SWAP_FLAG_PREFER
)
2088 (swap_flags
& SWAP_FLAG_PRIO_MASK
) >> SWAP_FLAG_PRIO_SHIFT
;
2089 enable_swap_info(p
, prio
, swap_map
);
2091 printk(KERN_INFO
"Adding %uk swap on %s. "
2092 "Priority:%d extents:%d across:%lluk %s%s\n",
2093 p
->pages
<<(PAGE_SHIFT
-10), name
, p
->prio
,
2094 nr_extents
, (unsigned long long)span
<<(PAGE_SHIFT
-10),
2095 (p
->flags
& SWP_SOLIDSTATE
) ? "SS" : "",
2096 (p
->flags
& SWP_DISCARDABLE
) ? "D" : "");
2098 mutex_unlock(&swapon_mutex
);
2099 atomic_inc(&proc_poll_event
);
2100 wake_up_interruptible(&proc_poll_wait
);
2102 if (S_ISREG(inode
->i_mode
))
2103 inode
->i_flags
|= S_SWAPFILE
;
2107 if (inode
&& S_ISBLK(inode
->i_mode
) && p
->bdev
) {
2108 set_blocksize(p
->bdev
, p
->old_block_size
);
2109 blkdev_put(p
->bdev
, FMODE_READ
| FMODE_WRITE
| FMODE_EXCL
);
2111 destroy_swap_extents(p
);
2112 swap_cgroup_swapoff(p
->type
);
2113 spin_lock(&swap_lock
);
2114 p
->swap_file
= NULL
;
2116 spin_unlock(&swap_lock
);
2119 if (inode
&& S_ISREG(inode
->i_mode
)) {
2120 mutex_unlock(&inode
->i_mutex
);
2123 filp_close(swap_file
, NULL
);
2126 if (page
&& !IS_ERR(page
)) {
2128 page_cache_release(page
);
2132 if (inode
&& S_ISREG(inode
->i_mode
))
2133 mutex_unlock(&inode
->i_mutex
);
2137 void si_swapinfo(struct sysinfo
*val
)
2140 unsigned long nr_to_be_unused
= 0;
2142 spin_lock(&swap_lock
);
2143 for (type
= 0; type
< nr_swapfiles
; type
++) {
2144 struct swap_info_struct
*si
= swap_info
[type
];
2146 if ((si
->flags
& SWP_USED
) && !(si
->flags
& SWP_WRITEOK
))
2147 nr_to_be_unused
+= si
->inuse_pages
;
2149 val
->freeswap
= nr_swap_pages
+ nr_to_be_unused
;
2150 val
->totalswap
= total_swap_pages
+ nr_to_be_unused
;
2151 spin_unlock(&swap_lock
);
2155 * Verify that a swap entry is valid and increment its swap map count.
2157 * Returns error code in following case.
2159 * - swp_entry is invalid -> EINVAL
2160 * - swp_entry is migration entry -> EINVAL
2161 * - swap-cache reference is requested but there is already one. -> EEXIST
2162 * - swap-cache reference is requested but the entry is not used. -> ENOENT
2163 * - swap-mapped reference requested but needs continued swap count. -> ENOMEM
2165 static int __swap_duplicate(swp_entry_t entry
, unsigned char usage
)
2167 struct swap_info_struct
*p
;
2168 unsigned long offset
, type
;
2169 unsigned char count
;
2170 unsigned char has_cache
;
2173 if (non_swap_entry(entry
))
2176 type
= swp_type(entry
);
2177 if (type
>= nr_swapfiles
)
2179 p
= swap_info
[type
];
2180 offset
= swp_offset(entry
);
2182 spin_lock(&swap_lock
);
2183 if (unlikely(offset
>= p
->max
))
2186 count
= p
->swap_map
[offset
];
2187 has_cache
= count
& SWAP_HAS_CACHE
;
2188 count
&= ~SWAP_HAS_CACHE
;
2191 if (usage
== SWAP_HAS_CACHE
) {
2193 /* set SWAP_HAS_CACHE if there is no cache and entry is used */
2194 if (!has_cache
&& count
)
2195 has_cache
= SWAP_HAS_CACHE
;
2196 else if (has_cache
) /* someone else added cache */
2198 else /* no users remaining */
2201 } else if (count
|| has_cache
) {
2203 if ((count
& ~COUNT_CONTINUED
) < SWAP_MAP_MAX
)
2205 else if ((count
& ~COUNT_CONTINUED
) > SWAP_MAP_MAX
)
2207 else if (swap_count_continued(p
, offset
, count
))
2208 count
= COUNT_CONTINUED
;
2212 err
= -ENOENT
; /* unused swap entry */
2214 p
->swap_map
[offset
] = count
| has_cache
;
2217 spin_unlock(&swap_lock
);
2222 printk(KERN_ERR
"swap_dup: %s%08lx\n", Bad_file
, entry
.val
);
2227 * Help swapoff by noting that swap entry belongs to shmem/tmpfs
2228 * (in which case its reference count is never incremented).
2230 void swap_shmem_alloc(swp_entry_t entry
)
2232 __swap_duplicate(entry
, SWAP_MAP_SHMEM
);
2236 * Increase reference count of swap entry by 1.
2237 * Returns 0 for success, or -ENOMEM if a swap_count_continuation is required
2238 * but could not be atomically allocated. Returns 0, just as if it succeeded,
2239 * if __swap_duplicate() fails for another reason (-EINVAL or -ENOENT), which
2240 * might occur if a page table entry has got corrupted.
2242 int swap_duplicate(swp_entry_t entry
)
2246 while (!err
&& __swap_duplicate(entry
, 1) == -ENOMEM
)
2247 err
= add_swap_count_continuation(entry
, GFP_ATOMIC
);
2252 * @entry: swap entry for which we allocate swap cache.
2254 * Called when allocating swap cache for existing swap entry,
2255 * This can return error codes. Returns 0 at success.
2256 * -EBUSY means there is a swap cache.
2257 * Note: return code is different from swap_duplicate().
2259 int swapcache_prepare(swp_entry_t entry
)
2261 return __swap_duplicate(entry
, SWAP_HAS_CACHE
);
2265 * add_swap_count_continuation - called when a swap count is duplicated
2266 * beyond SWAP_MAP_MAX, it allocates a new page and links that to the entry's
2267 * page of the original vmalloc'ed swap_map, to hold the continuation count
2268 * (for that entry and for its neighbouring PAGE_SIZE swap entries). Called
2269 * again when count is duplicated beyond SWAP_MAP_MAX * SWAP_CONT_MAX, etc.
2271 * These continuation pages are seldom referenced: the common paths all work
2272 * on the original swap_map, only referring to a continuation page when the
2273 * low "digit" of a count is incremented or decremented through SWAP_MAP_MAX.
2275 * add_swap_count_continuation(, GFP_ATOMIC) can be called while holding
2276 * page table locks; if it fails, add_swap_count_continuation(, GFP_KERNEL)
2277 * can be called after dropping locks.
2279 int add_swap_count_continuation(swp_entry_t entry
, gfp_t gfp_mask
)
2281 struct swap_info_struct
*si
;
2284 struct page
*list_page
;
2286 unsigned char count
;
2289 * When debugging, it's easier to use __GFP_ZERO here; but it's better
2290 * for latency not to zero a page while GFP_ATOMIC and holding locks.
2292 page
= alloc_page(gfp_mask
| __GFP_HIGHMEM
);
2294 si
= swap_info_get(entry
);
2297 * An acceptable race has occurred since the failing
2298 * __swap_duplicate(): the swap entry has been freed,
2299 * perhaps even the whole swap_map cleared for swapoff.
2304 offset
= swp_offset(entry
);
2305 count
= si
->swap_map
[offset
] & ~SWAP_HAS_CACHE
;
2307 if ((count
& ~COUNT_CONTINUED
) != SWAP_MAP_MAX
) {
2309 * The higher the swap count, the more likely it is that tasks
2310 * will race to add swap count continuation: we need to avoid
2311 * over-provisioning.
2317 spin_unlock(&swap_lock
);
2322 * We are fortunate that although vmalloc_to_page uses pte_offset_map,
2323 * no architecture is using highmem pages for kernel pagetables: so it
2324 * will not corrupt the GFP_ATOMIC caller's atomic pagetable kmaps.
2326 head
= vmalloc_to_page(si
->swap_map
+ offset
);
2327 offset
&= ~PAGE_MASK
;
2330 * Page allocation does not initialize the page's lru field,
2331 * but it does always reset its private field.
2333 if (!page_private(head
)) {
2334 BUG_ON(count
& COUNT_CONTINUED
);
2335 INIT_LIST_HEAD(&head
->lru
);
2336 set_page_private(head
, SWP_CONTINUED
);
2337 si
->flags
|= SWP_CONTINUED
;
2340 list_for_each_entry(list_page
, &head
->lru
, lru
) {
2344 * If the previous map said no continuation, but we've found
2345 * a continuation page, free our allocation and use this one.
2347 if (!(count
& COUNT_CONTINUED
))
2350 map
= kmap_atomic(list_page
) + offset
;
2355 * If this continuation count now has some space in it,
2356 * free our allocation and use this one.
2358 if ((count
& ~COUNT_CONTINUED
) != SWAP_CONT_MAX
)
2362 list_add_tail(&page
->lru
, &head
->lru
);
2363 page
= NULL
; /* now it's attached, don't free it */
2365 spin_unlock(&swap_lock
);
2373 * swap_count_continued - when the original swap_map count is incremented
2374 * from SWAP_MAP_MAX, check if there is already a continuation page to carry
2375 * into, carry if so, or else fail until a new continuation page is allocated;
2376 * when the original swap_map count is decremented from 0 with continuation,
2377 * borrow from the continuation and report whether it still holds more.
2378 * Called while __swap_duplicate() or swap_entry_free() holds swap_lock.
2380 static bool swap_count_continued(struct swap_info_struct
*si
,
2381 pgoff_t offset
, unsigned char count
)
2387 head
= vmalloc_to_page(si
->swap_map
+ offset
);
2388 if (page_private(head
) != SWP_CONTINUED
) {
2389 BUG_ON(count
& COUNT_CONTINUED
);
2390 return false; /* need to add count continuation */
2393 offset
&= ~PAGE_MASK
;
2394 page
= list_entry(head
->lru
.next
, struct page
, lru
);
2395 map
= kmap_atomic(page
) + offset
;
2397 if (count
== SWAP_MAP_MAX
) /* initial increment from swap_map */
2398 goto init_map
; /* jump over SWAP_CONT_MAX checks */
2400 if (count
== (SWAP_MAP_MAX
| COUNT_CONTINUED
)) { /* incrementing */
2402 * Think of how you add 1 to 999
2404 while (*map
== (SWAP_CONT_MAX
| COUNT_CONTINUED
)) {
2406 page
= list_entry(page
->lru
.next
, struct page
, lru
);
2407 BUG_ON(page
== head
);
2408 map
= kmap_atomic(page
) + offset
;
2410 if (*map
== SWAP_CONT_MAX
) {
2412 page
= list_entry(page
->lru
.next
, struct page
, lru
);
2414 return false; /* add count continuation */
2415 map
= kmap_atomic(page
) + offset
;
2416 init_map
: *map
= 0; /* we didn't zero the page */
2420 page
= list_entry(page
->lru
.prev
, struct page
, lru
);
2421 while (page
!= head
) {
2422 map
= kmap_atomic(page
) + offset
;
2423 *map
= COUNT_CONTINUED
;
2425 page
= list_entry(page
->lru
.prev
, struct page
, lru
);
2427 return true; /* incremented */
2429 } else { /* decrementing */
2431 * Think of how you subtract 1 from 1000
2433 BUG_ON(count
!= COUNT_CONTINUED
);
2434 while (*map
== COUNT_CONTINUED
) {
2436 page
= list_entry(page
->lru
.next
, struct page
, lru
);
2437 BUG_ON(page
== head
);
2438 map
= kmap_atomic(page
) + offset
;
2445 page
= list_entry(page
->lru
.prev
, struct page
, lru
);
2446 while (page
!= head
) {
2447 map
= kmap_atomic(page
) + offset
;
2448 *map
= SWAP_CONT_MAX
| count
;
2449 count
= COUNT_CONTINUED
;
2451 page
= list_entry(page
->lru
.prev
, struct page
, lru
);
2453 return count
== COUNT_CONTINUED
;
2458 * free_swap_count_continuations - swapoff free all the continuation pages
2459 * appended to the swap_map, after swap_map is quiesced, before vfree'ing it.
2461 static void free_swap_count_continuations(struct swap_info_struct
*si
)
2465 for (offset
= 0; offset
< si
->max
; offset
+= PAGE_SIZE
) {
2467 head
= vmalloc_to_page(si
->swap_map
+ offset
);
2468 if (page_private(head
)) {
2469 struct list_head
*this, *next
;
2470 list_for_each_safe(this, next
, &head
->lru
) {
2472 page
= list_entry(this, struct page
, lru
);